The US Census provides a wealth of detailed information about the population, but its packaged data lacks sufficient attributes to answer certain questions. Machine learning can help overcome this limitation by upsampling the data to finer spatial resolutions, such as Census Block Groups. By using a predictive model trained on PUMAS data at a coarser scale, the model can be applied to the Census Block Group level to create new variables and produce maps of specific populations, like those with vision and hearing problems. This approach enables segmentation, which is valuable for various applications, including non-profits, businesses, and election campaigns.